2.5 Modality-Aware Rubrics
Study Time: 35 minutes Prerequisites: Module 2.4 (Objectivity & Thresholds) Learning Objectives:
- Integrate all five core properties into complete rubrics
- Build modality-specific criteria for audio and video tasks
- Build a submission checklist from any project's guidelines, and apply it before submitting
- Handle multimodal tasks with cross-modality integration
- Create platform-ready evaluation rubrics
Introduction
This module integrates everything you've learned about rubric engineering. You'll combine all five core properties (atomic, instance-specific, self-contained, objective, modality-aware) to build complete, certification-ready rubrics.
The Five Core Properties:
- Atomic (Module 2.1): Tests exactly ONE aspect
- Instance-Specific (Module 2.2): References actual task content
- Self-Contained (Module 2.3): Stands alone without cross-referencing
- Objective (Module 2.4): Includes measurable thresholds
- Modality-Aware (This module): Evaluates modality-specific dimensions
2.5.1 Building a Submission Checklist
Why you build this yourself
Every project defines quality slightly differently. A checklist you were handed covers the properties someone else's project cared about. The durable skill is converting any set of project guidelines into a list you can run before submitting.
The method
- Read the guidelines end to end before writing anything. Skimming for the parts that look relevant is how requirements get missed.
- Extract every stated requirement. Note anything phrased as must, should, always, never, or a stated format. Include the negatives.
- Convert each into a binary check. One you can answer yes or no in seconds. "Is the tone appropriate?" is not a check. "Does every criterion state a measurable threshold?" is.
- Cut anything you cannot verify by looking. If answering it means re-doing the analysis, it belongs in your process, not your final pass.
- Order by cost of being wrong. Whatever fails a submission outright goes first.
Worked example: the five core properties
Applying that method to a project requiring atomic, instance-specific, self-contained, objective, and modality-aware criteria produces the checklist below. Treat it as one instance of the method, not the universal list. The properties you check can differ from project to project, and the project's guidelines are always authoritative.
For Each Criterion (verify which properties your project requires):
1. Atomic?
- Tests exactly ONE aspect
- No "and" connecting different things
- Passes failure scenario test (can't pass one part and fail another)
2. Instance-Specific?
- References actual content from THIS task
- Names specific facts, terms, numbers, components
- Can't be copy-pasted to a different task
3. Self-Contained?
- Understood without reading other criteria
- No "as mentioned above" or "see criterion X"
- Includes all needed context
4. Objective?
- Includes measurable threshold or standard
- No undefined "somewhat," "fairly," "adequate"
- Two evaluators would agree on the answer
5. Modality-Aware? (when specified by project)
- Audio tasks: may require clarity, pace, tone, or quality criteria
- Video tasks: may require visual and audio criteria
- Multimodal: may require integration criteria
The hands-on part starts here
Unlock the full lesson
- The step-by-step evaluation framework
- Graded practice drills with instant feedback
- Full video walkthrough
- Kappa, your AI study partner, for guided practice
- Downloadable rubric templates
- Module checkpoint quiz